Question: field_validator Pydantic and ModelRetry Pydantic AI

[YanSte]

Question

Hi Pydantic AI team,

I couldn’t find documentation about using field_validator together with ModelRetry.

Does using field_validator with ModelRetry actually work, and will it cause the LLM to retry with the updated sentence?

Or should I use ValueError?

Thanks in advance.

Additional Context

Here’s my case example:

@final
class Task(BaseModel):
    """A task in an execution plan."""

    task_name: str = Field(
        description='Name of the specialized task to execute.',
    )

    @field_validator('task_name', mode='before')
    @classmethod
    def _upper_task_name(cls, value: str):
        value = value.strip()
        if not value:
            raise ModelRetry('task_name is required and cannot be empty.')

        return value.upper()



# OR

    @field_validator('task_name', mode='before')
    @classmethod
    def _upper_task_name(cls, value: str):
        value = value.strip()
        if not value:
            raise ValueError('task_name is required and cannot be empty.')

        return value.upper()

# Code

agent = Agent(
...
output_type=Task,
)

[YanSte]

@DouweM Maybe?

[DouweM]

If the field_validator fails because of a ValueError, Pydantic’s model_validate will raise a ValidationError, which Pydantic AI will handle like a ModelRetry and send it back to the model. You should be able to confirm this using ai.pydantic.dev/logfire